ABSTRACT We use the SOUX sample of ∼700 active galactic nucleus (AGN) to form average optical-ultraviolet (UV)-X-rays spectral energy distributions (SEDs) on a two-dimensional (2D) grid of MBH and L2500. We compare these with the predictions of a new AGN SED model, qsosed, which includes prescriptions for both hot and warm Comptonization regions as well as an outer standard disc. This predicts the overall SED fairly well for 7.5 < log(MBH/M⊙) < 9.0 over a wide range in L/LEdd, but at higher masses the outer disc spectra in the model are far too cool to match the data. We create optical-UV composites from the entire Sloan Digital Sky Survey sample and use these to show that the mismatch is due to there being no significant change in spectral shape of the optical-UV continuum across several decades of MBH at constant luminosity. We show for the first time that this cannot be matched by standard disc models with high black hole spin. These apparently fit, but are not self-consistent as they do not include the General Relativistic effects for the emission to reach the observer. At high spin, increased gravitational redshift compensates for almost all of the higher temperature emission from the smaller inner disc radii. The data do not match the predictions made by any current accretion flow model. Either the disc is completely covered by a warm Comptonization layer whose properties change systematically with L/LEdd, or the accretion flow structure is fundamentally different to that of the standard disc models.